Probabilistic Forecasts of Daily Reference Evapotranspiration for the Continental U.S. Based on Numerical Weather Predictions.

Predicting evapotranspiration is fundamental in hydrological applications for addressing water resources and irrigation management issues. Numerical weather prediction outputs combined with modern statistical post-processing techniques such as the Ensemble Model Output Statistics (EMOS) can be used for producing medium range (1-10 days) reference evapotranspiration (ETo) forecasts at daily timescale. This work is aimed to evaluate the performance of raw and EMOS post-processed medium-range ETo forecasts over the continental U.S. using the European Centre for Medium-Range Weather Forecasts (ECMWF) perturbed ensemble forecasts as basic inputs. The effect of the size of the ensemble, the estimation method of ETo as well as the length of the training period on several deterministic and probabilistic performance statistics is pondered as part of the study. ETo estimated based on the FAO 56 Penman Monteith equation with the quality-controlled U.S. Regional Climate Reference Network measurements are adopted as observational reference. The results show that the application of the EMOS post processing technique improves the raw forecast as indicated by most of the performance statistics, with the rate of the improvements being noticeably higher for the largest lead times. However, the performance of the calibrated forecasts was found scarcely sensitive to the adopted lengths of the training period. Insights about the more efficient ETo models according to the region are provided. Our results demonstrate the potential for using the adopted framework in operational evapotranspiration and irrigation advisory systems at national scale.